feat: add google vertex ai (claude) provider by SadriG91 · Pull Request #179 · langchain-ai/openwiki

SadriG91 · 2026-07-07T10:41:57Z

What

Adds a vertex provider that runs Anthropic's Claude models through Google Vertex AI, authenticated with Google Application Default Credentials — no API key:

OPENWIKI_PROVIDER=vertex
GOOGLE_CLOUD_PROJECT=your-gcp-project
GOOGLE_CLOUD_LOCATION=global   # optional, defaults to global

Auth works through the standard ADC chain: GOOGLE_APPLICATION_CREDENTIALS (service-account key), gcloud auth application-default login, or workload identity.

Why

Organizations on Google Cloud consume Claude through Vertex AI (billing, quotas, IAM, data residency) rather than a direct Anthropic API key — today OpenWiki can't use those deployments. No existing issue or PR covers Claude-on-Vertex (#154/#36 add Gemini models, a different provider path).

How

Reuses the existing ChatAnthropic class with its documented createClient option returning an AnthropicVertex client from @anthropic-ai/vertex-sdk — the same message/tool-calling code path as the anthropic provider; only authentication differs.
ProviderConfig.apiKeyEnvKey becomes optional. A new getMissingProviderEnvKey() helper powers the agent's credential gate, both CLI non-interactive gates, and needsCredentialSetup(), so key-less providers are handled uniformly (with an ADC hint appended to error messages).
Interactive onboarding: vertex skips the API-key step and instead prompts for GOOGLE_CLOUD_PROJECT (required) and GOOGLE_CLOUD_LOCATION (optional; Enter keeps the soft global default — nothing persisted).
GOOGLE_CLOUD_PROJECT / GOOGLE_CLOUD_LOCATION / GOOGLE_APPLICATION_CREDENTIALS join MANAGED_ENV_KEYS and show as non-secret values in credential diagnostics.
isValidModelId now accepts @ so Vertex's versioned IDs validate (claude-haiku-4-5@20251001); Sonnet/Opus use the dateless IDs.
The invalid-provider message in cli.tsx is now derived from SELECTABLE_OPENWIKI_PROVIDERS (the hardcoded list had gone stale — it was missing openai-compatible). Heads-up: feat: add gemini (ai studio) and gemini enterprise (vertex ai) providers with vertex model-family routing #154 makes the same tidy-up, so whichever lands second has a trivial conflict there.

How I tested it

pnpm run format, pnpm run lint, and pnpm test all pass (98 tests, including new coverage for the vertex config, the getMissingProviderEnvKey/resolveProviderLocation/providerRequiresApiKey helpers, @-versioned model IDs through both isValidModelId and --modelId parsing, and the managed env keys).
Non-interactive gate: running with OPENWIKI_PROVIDER=vertex and no GOOGLE_CLOUD_PROJECT exits 1 naming the env key plus the ADC hint.
Interactive onboarding walked end-to-end in a sandbox HOME: no API-key step, required project step (rejects empty/whitespace input), optional location step, the three Claude presets in the model menu, and correct ~/.openwiki/.env contents afterwards.
Real end-to-end: one-shot and full-agent runs against a live GCP project using gcloud ADC on the global endpoint, with claude-haiku-4-5@20251001 responding through the deepagents tool-calling path.

README and the repo's openwiki/ docs are updated to match (same file set as the openai-compatible provider PR, d75a9f3).

🤖 Generated with Claude Code

SadriG91 · 2026-07-07T15:18:34Z

Heads-up on some convergence: the AWS Bedrock PRs (#185, #96, and #40 for issue #56) generalize the same thing this PR does for Vertex — credential gating for providers that don't authenticate with a single API key.

feat: add AWS Bedrock as a model provider #185 models it as an explicit auth: "api-key" | "aws" discriminant on the provider config, with region-aware validation.
This PR models it by making apiKeyEnvKey optional and deriving the gate from whichever required env keys the provider declares (getMissingProviderEnvKey()), with a per-provider credential hint for the error paths.

The two shapes enable the same thing, so whichever pattern maintainers prefer for keyless/cloud-credential providers, this change adapts trivially — happy to rebase onto the Bedrock abstraction if one of those lands first (or they can rebase onto this one; the vertex entry would translate to auth: "gcp" in a one-line change). Also flagging #154 (Gemini), which shares the GOOGLE_CLOUD_PROJECT/GOOGLE_CLOUD_LOCATION env keys and makes the same SELECTABLE_OPENWIKI_PROVIDERS-derived error-message tidy-up in cli.tsx.

Brad Huffman (ppsplus-bradh) · 2026-07-12T19:35:45Z

SadriG91 nice PR — we independently landed on the exact same ChatAnthropic + createClient → AnthropicVertex bridge, which is reassuring. Flagging the overlap and offering to consolidate so the maintainers don't have to reconcile two Vertex code paths.

How it overlaps with #154: #154 (Gemini + Gemini Enterprise) also does Claude-on-Vertex, but as one part of a broader gemini-enterprise provider that routes by model family: Gemini/Gemma → ChatGoogle (native generateContent), Claude → AnthropicVertex (rawPredict), and Llama/Mistral/DeepSeek/Qwen → the Vertex OpenAI-compatible endpoint — dispatched from the model ID via resolveVertexSurface(). So #179's Claude support is essentially a subset of #154's routing, and both carry the same @-in-isValidModelId fix and the SELECTABLE_OPENWIKI_PROVIDERS tidy-up you already flagged as a conflict.

Where your PR is cleaner: the key-less-provider abstraction (getMissingProviderEnvKey / providerRequiresApiKey / optional apiKeyEnvKey). #154 currently leans on the GOOGLE_CLOUD_PROJECT-as-apiKey shortcut for gemini-enterprise, which is hackier than your explicit handling — I'd happily adopt yours.

One heads-up that affects vertex too: AnthropicVertex extends the base Anthropic SDK, which reads ANTHROPIC_API_KEY / ANTHROPIC_AUTH_TOKEN from the environment and sends them as an Authorization header, clobbering the Google OAuth token (ACCESS_TOKEN_TYPE_UNSUPPORTED). It bites users who previously configured the anthropic provider and left the key in ~/.openwiki/.env. #154 neutralizes those two vars around the (synchronous) constructor — worth guarding in #179 too if it lands standalone.

Offer: happy to consolidate whichever direction the maintainers prefer — either fold your key-less-provider abstraction into #154's model-family routing, or rebase #154's Gemini + MaaS routing on top of #179. I'll do the merge work either way. Deferring to the maintainers on the naming question too (vertex vs gemini-enterprise, or a single Vertex provider that routes all families).

SadriG91 · 2026-07-13T09:05:24Z

Thanks Brad Huffman (@ppsplus-bradh) — this is exactly the kind of cross-PR review this repo needs right now, and your credentials heads-up was a real catch.

Verified and fixed. I reproduced it against a live Vertex project: a stray ANTHROPIC_AUTH_TOKEN fails Vertex auth exactly as you describe (the base SDK captures it and wins the Authorization header); interestingly a stray ANTHROPIC_API_KEY alone didn't break in my environment — but since that's SDK-version-dependent luck, a06156b now masks both around the (synchronous) AnthropicVertex construction and restores them after, same approach as yours. Both scenarios re-tested live: working.

On consolidation: I'd take you up on the second direction — land #179 first, then rebase #154's Gemini + MaaS family routing on top of it. Reasoning: this repo's CONTRIBUTING pushes hard for one-change PRs, and #179 is the smallest reviewable unit (the key-less-provider abstraction + the Claude path, which has now had multiple live end-to-end runs including a full --init on a production repo). Multi-family routing via resolveVertexSurface() is genuinely useful but a bigger design decision the maintainers should get to evaluate on its own. The abstraction is yours to adopt either way — glad it's useful.

Naming: happy to defer to maintainers. One consideration for them: a vertex provider can grow family routing behind the same name later without breaking anyone, whereas gemini-enterprise serving Claude/Llama/Mistral may read surprisingly in configs. But either works with either consolidation direction.

@SadriG91

The blank-region default of us-central1 returns 400 'Model not servable in region' for Anthropic Claude on Vertex. 'global' serves both Gemini and the Claude surface. MaaS partner models remain region-specific, so those users set GOOGLE_CLOUD_LOCATION explicitly. Flagged by @SadriG91 on langchain-ai#154 (also addressed in langchain-ai#179). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Colin Francis (colifran) · 2026-07-15T02:20:59Z

SadriG91 - thanks for the PR! Taking a look.

Colin Francis (colifran) · 2026-07-15T02:40:54Z

  const retryOptions = { maxRetries: providerRetryAttempts };

+  if (provider === "vertex") {
+    return new ChatAnthropic(modelId, {


i think we need to pass retryOptions here

Good catch — added, now 55d868f after rebasing over the Bedrock/Nebius merges, so vertex honors OPENWIKI_PROVIDER_RETRY_ATTEMPTS like the other providers. Re-ran the full suite (typecheck, lint, 260 tests) and a live smoke test against Vertex on the global endpoint — all green.

Colin Francis (colifran)

Looks great SadriG91! I think we need to pass in retryOptions but good to merge after that. I have a PR open to fix the audit issue in CI.

Adds a 'vertex' provider that runs Anthropic Claude models through Google Vertex AI, authenticated with Application Default Credentials (no API key): ChatAnthropic with a custom AnthropicVertex client (@anthropic-ai/vertex-sdk) via the documented createClient option. - constants: vertex provider entry (Haiku/Sonnet/Opus on Vertex), apiKeyEnvKey made optional, new getMissingProviderEnvKey / resolveProviderLocation / getProviderCredentialHint helpers; isValidModelId now accepts @-versioned Vertex model IDs - agent: vertex branch in createModel; ensureProviderKey generalized to ensureProviderCredentials - onboarding: GCP project (required) and location (optional, defaults to global) steps replace the API-key step for vertex - env: GOOGLE_CLOUD_PROJECT / GOOGLE_CLOUD_LOCATION / GOOGLE_APPLICATION_CREDENTIALS managed and shown in diagnostics - cli: credential gates report the missing env key with an ADC hint; invalid-provider message now derived from SELECTABLE_OPENWIKI_PROVIDERS - tests, README, and openwiki/ docs updated

…ient The Anthropic SDK captures ANTHROPIC_API_KEY / ANTHROPIC_AUTH_TOKEN from the environment at client construction and sends them as request credentials, displacing the Google OAuth token (reproduced: a stray ANTHROPIC_AUTH_TOKEN fails Vertex auth — e.g. left over from a previous anthropic-provider setup). Construct AnthropicVertex with both masked, then restore them.

SadriG91 · 2026-07-15T11:34:40Z

Thanks for the review Colin Francis (@colifran)! Two updates since:

retryOptions — done in the commit I replied with inline; vertex now honors OPENWIKI_PROVIDER_RETRY_ATTEMPTS like every other provider.
Rebased onto current main (now 55d868f) — the Bedrock (feat: add aws bedrock to the model provider list #327) and Nebius (feat: add nebius token factory provider support #147) providers landed since your review and touched the same provider surface, so I reconciled: Bedrock's secret-key/region onboarding steps and vertex's GCP project/location steps now coexist in the setup wizard and step chain, and Bedrock's accessKeyId lookup goes through the same optional-apiKeyEnvKey helpers this PR introduces. Full suite passes (typecheck, lint, 260 tests) plus a live end-to-end run against Vertex on the global endpoint.

Saw the Trivy audit fix landed too (#338), so CI should be fully green on re-run. Ready when you are.

Colin Francis (colifran)

Great job! Thanks SadriG91!

…enterprise Add two Google model providers and route the Vertex AI (now Gemini Enterprise) Model Garden through a single provider: - `gemini` (AI Studio): Google's Gemini models via one API key (GEMINI_API_KEY). - `gemini-enterprise` (Vertex AI): keyless, ADC-authenticated provider that routes each model ID to the right Model Garden surface — native Gemini/Gemma over generateContent, Claude over the Anthropic Vertex SDK, and partner/open-weight models over the OpenAI-compatible MaaS endpoint (resolveVertexSurface / createGeminiEnterpriseModel). This supersedes the discrete `vertex` (Claude-only) provider added in langchain-ai#179: `gemini-enterprise` reaches Claude on Vertex through the same AnthropicVertex construction and ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN env-neutralization, plus publisher-path stripping, so no capability is lost. The rename tracks Google's Cloud Next 2026 rebrand of Vertex AI to the Gemini Enterprise Agent Platform while keeping the unchanged underlying API surface. Builds on langchain-ai#179's keyless-provider plumbing (projectEnvKey/locationEnvKey/ defaultLocation, getMissingProviderEnvKey, resolveProviderLocation, getProviderCredentialHint) and its gcp-project/gcp-location onboarding steps, which gemini-enterprise inherits — dropping langchain-ai#154's earlier approach of overloading apiKeyEnvKey to carry the non-secret GOOGLE_CLOUD_PROJECT. Gemini uses disableStreaming + outputVersion "v0" to preserve Gemini 3.x thought signatures across multi-turn tool calls (see the comment in createGeminiEnterpriseModel). Co-Authored-By: Claude <noreply@anthropic.com>

…enterprise Add two Google model providers and route the Vertex AI (now Gemini Enterprise) Model Garden through a single provider: - `gemini` (AI Studio): Google's Gemini models via one API key (GEMINI_API_KEY). - `gemini-enterprise` (Vertex AI): keyless, ADC-authenticated provider that routes each model ID to the right Model Garden surface — native Gemini/Gemma over generateContent, Claude over the Anthropic Vertex SDK, and partner/open-weight models over the OpenAI-compatible MaaS endpoint (resolveVertexSurface / createGeminiEnterpriseModel). This supersedes the discrete `vertex` (Claude-only) provider added in langchain-ai#179: `gemini-enterprise` reaches Claude on Vertex through the same AnthropicVertex construction and ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN env-neutralization, plus publisher-path stripping, so no capability is lost. The rename tracks Google's Cloud Next 2026 rebrand of Vertex AI to the Gemini Enterprise Agent Platform while keeping the unchanged underlying API surface. Builds on langchain-ai#179's keyless-provider plumbing (projectEnvKey/locationEnvKey/ defaultLocation, getMissingProviderEnvKey, resolveProviderLocation, getProviderCredentialHint) and its gcp-project/gcp-location onboarding steps, which gemini-enterprise inherits — dropping langchain-ai#154's earlier approach of overloading apiKeyEnvKey to carry the non-secret GOOGLE_CLOUD_PROJECT. Gemini uses disableStreaming + outputVersion "v0" to preserve Gemini 3.x thought signatures across multi-turn tool calls (shared GEMINI_THOUGHT_SIGNATURE_OPTIONS across both ChatGoogle surfaces). The MaaS endpoint and the Gemini surface both special-case the `global` location: MaaS uses the unprefixed aiplatform.googleapis.com host, and the Gemini surface strips a publisher-pathed ID to the bare model the generateContent surface expects. resolveVertexSurface lists `codellama` explicitly so bare codellama-* IDs route to MaaS. Co-Authored-By: Claude <noreply@anthropic.com>

@avidspartan1

…ers with vertex model-family routing (#154) * feat: add Gemini providers and consolidate Vertex Claude into gemini-enterprise Add two Google model providers and route the Vertex AI (now Gemini Enterprise) Model Garden through a single provider: - `gemini` (AI Studio): Google's Gemini models via one API key (GEMINI_API_KEY). - `gemini-enterprise` (Vertex AI): keyless, ADC-authenticated provider that routes each model ID to the right Model Garden surface — native Gemini/Gemma over generateContent, Claude over the Anthropic Vertex SDK, and partner/open-weight models over the OpenAI-compatible MaaS endpoint (resolveVertexSurface / createGeminiEnterpriseModel). This supersedes the discrete `vertex` (Claude-only) provider added in #179: `gemini-enterprise` reaches Claude on Vertex through the same AnthropicVertex construction and ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN env-neutralization, plus publisher-path stripping, so no capability is lost. The rename tracks Google's Cloud Next 2026 rebrand of Vertex AI to the Gemini Enterprise Agent Platform while keeping the unchanged underlying API surface. Builds on #179's keyless-provider plumbing (projectEnvKey/locationEnvKey/ defaultLocation, getMissingProviderEnvKey, resolveProviderLocation, getProviderCredentialHint) and its gcp-project/gcp-location onboarding steps, which gemini-enterprise inherits — dropping #154's earlier approach of overloading apiKeyEnvKey to carry the non-secret GOOGLE_CLOUD_PROJECT. Gemini uses disableStreaming + outputVersion "v0" to preserve Gemini 3.x thought signatures across multi-turn tool calls (shared GEMINI_THOUGHT_SIGNATURE_OPTIONS across both ChatGoogle surfaces). The MaaS endpoint and the Gemini surface both special-case the `global` location: MaaS uses the unprefixed aiplatform.googleapis.com host, and the Gemini surface strips a publisher-pathed ID to the bare model the generateContent surface expects. resolveVertexSurface lists `codellama` explicitly so bare codellama-* IDs route to MaaS. Co-Authored-By: Claude <noreply@anthropic.com> * test: verify provider retry attempts reach the Gemini constructors createModel now spreads ...retryOptions into every Gemini surface (the AI Studio ChatGoogle and all three gemini-enterprise clients), which previously omitted it — so OPENWIKI_PROVIDER_RETRY_ATTEMPTS was silently ignored and a 429 failed fast instead of backing off (reported by @avidspartan1 on #154). maxRetries is not a readable property on the constructed LangChain models, so this asserts it via vi.mock capturing each constructor's options (including that maxRetries: 0 propagates and the AI Studio API key reaches ChatGoogle). Once maxRetries is wired through, @langchain/core's AsyncCaller already honors a server-provided Retry-After (up to 60s), so the existing backoff engages. Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>

@avidspartan1

…fixes fix (#357) * ci: fix repository posture checks * feat: Add trending repo badge to readme (#347) * chore(deps): bump the major group with 5 updates (#348) Bumps the major group with 5 updates: | Package | From | To | | --- | --- | --- | | [actions/checkout](https://github.com/actions/checkout) | `4` | `7` | | [pnpm/action-setup](https://github.com/pnpm/action-setup) | `4.3.0` | `6.0.9` | | [actions/setup-node](https://github.com/actions/setup-node) | `4` | `7` | | [github/codeql-action](https://github.com/github/codeql-action) | `3` | `4` | | [peter-evans/create-pull-request](https://github.com/peter-evans/create-pull-request) | `7.0.11` | `8.1.1` | Updates `actions/checkout` from 4 to 7 - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](actions/checkout@v4...v7) Updates `pnpm/action-setup` from 4.3.0 to 6.0.9 - [Release notes](https://github.com/pnpm/action-setup/releases) - [Commits](pnpm/action-setup@b906aff...0ebf471) Updates `actions/setup-node` from 4 to 7 - [Release notes](https://github.com/actions/setup-node/releases) - [Commits](actions/setup-node@v4...v7) Updates `github/codeql-action` from 3 to 4 - [Release notes](https://github.com/github/codeql-action/releases) - [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md) - [Commits](github/codeql-action@v3...v4) Updates `peter-evans/create-pull-request` from 7.0.11 to 8.1.1 - [Release notes](https://github.com/peter-evans/create-pull-request/releases) - [Commits](peter-evans/create-pull-request@22a9089...5f6978f) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: pnpm/action-setup dependency-version: 6.0.9 dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: actions/setup-node dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: github/codeql-action dependency-version: '4' dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: peter-evans/create-pull-request dependency-version: 8.1.1 dependency-type: direct:production update-type: version-update:semver-major dependency-group: major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Colin Francis <131073567+colifran@users.noreply.github.com> * feat: add gemini (ai studio) and gemini enterprise (vertex ai) providers with vertex model-family routing (#154) * feat: add Gemini providers and consolidate Vertex Claude into gemini-enterprise Add two Google model providers and route the Vertex AI (now Gemini Enterprise) Model Garden through a single provider: - `gemini` (AI Studio): Google's Gemini models via one API key (GEMINI_API_KEY). - `gemini-enterprise` (Vertex AI): keyless, ADC-authenticated provider that routes each model ID to the right Model Garden surface — native Gemini/Gemma over generateContent, Claude over the Anthropic Vertex SDK, and partner/open-weight models over the OpenAI-compatible MaaS endpoint (resolveVertexSurface / createGeminiEnterpriseModel). This supersedes the discrete `vertex` (Claude-only) provider added in #179: `gemini-enterprise` reaches Claude on Vertex through the same AnthropicVertex construction and ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN env-neutralization, plus publisher-path stripping, so no capability is lost. The rename tracks Google's Cloud Next 2026 rebrand of Vertex AI to the Gemini Enterprise Agent Platform while keeping the unchanged underlying API surface. Builds on #179's keyless-provider plumbing (projectEnvKey/locationEnvKey/ defaultLocation, getMissingProviderEnvKey, resolveProviderLocation, getProviderCredentialHint) and its gcp-project/gcp-location onboarding steps, which gemini-enterprise inherits — dropping #154's earlier approach of overloading apiKeyEnvKey to carry the non-secret GOOGLE_CLOUD_PROJECT. Gemini uses disableStreaming + outputVersion "v0" to preserve Gemini 3.x thought signatures across multi-turn tool calls (shared GEMINI_THOUGHT_SIGNATURE_OPTIONS across both ChatGoogle surfaces). The MaaS endpoint and the Gemini surface both special-case the `global` location: MaaS uses the unprefixed aiplatform.googleapis.com host, and the Gemini surface strips a publisher-pathed ID to the bare model the generateContent surface expects. resolveVertexSurface lists `codellama` explicitly so bare codellama-* IDs route to MaaS. Co-Authored-By: Claude <noreply@anthropic.com> * test: verify provider retry attempts reach the Gemini constructors createModel now spreads ...retryOptions into every Gemini surface (the AI Studio ChatGoogle and all three gemini-enterprise clients), which previously omitted it — so OPENWIKI_PROVIDER_RETRY_ATTEMPTS was silently ignored and a 429 failed fast instead of backing off (reported by @avidspartan1 on #154). maxRetries is not a readable property on the constructed LangChain models, so this asserts it via vi.mock capturing each constructor's options (including that maxRetries: 0 propagates and the AI Studio API key reaches ChatGoogle). Once maxRetries is wired through, @langchain/core's AsyncCaller already honors a server-provided Retry-After (up to 60s), so the existing backoff engages. Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> * bump deepagents to 1.11.0 to restore execute tool via routePrefixes fix --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: John Kennedy <65985482+jkennedyvz@users.noreply.github.com> Co-authored-by: Brace Sproul <braceasproul@gmail.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Brad Huffman <brad.huffman@wellsky.com> Co-authored-by: Claude <noreply@anthropic.com>

@avidspartan1

* feat: Add support for OKF via prompting (#321) * feat: Add support for OKF via prompting * Apply suggestions from code review Co-authored-by: Brace Sproul <braceasproul@gmail.com> * cr * feat: OKF index.md files (#323) * feat: Add support for OKF via prompting * feat: OKF index.md files * Apply suggestions from code review Co-authored-by: Brace Sproul <braceasproul@gmail.com> * cr * cr * cr * docs: document index middleware helpers * Update src/agent/prompt.ts * cr * make interface * drop unnecc type * feat: Add OKF front matter validator (#324) * feat: Add support for OKF via prompting * feat: OKF index.md files * Apply suggestions from code review Co-authored-by: Brace Sproul <braceasproul@gmail.com> * cr * feat: Add OKF front matter validator * cr * cr * docs: document frontmatter validator helpers * docs: document index middleware helpers * Update src/agent/prompt.ts * cr * make interface * drop unnecc type * cr * cr * interfaces * cr * cr * feat: OKF update wiki skill (#339) * feat: Add support for OKF via prompting * feat: OKF index.md files * Apply suggestions from code review Co-authored-by: Brace Sproul <braceasproul@gmail.com> * cr * feat: Add OKF front matter validator * cr * cr * docs: document frontmatter validator helpers * docs: document index middleware helpers * Update src/agent/prompt.ts * cr * make interface * drop unnecc type * cr * cr * interfaces * cr * feat: OKF update wiki skill * update version * fix: Writing skills (#351) * fix: prompt agent to write one to two sentence desc instead of just one (#353) * fix: Better description prompting (#354) * fix: Better description prompting * cr * cr * feat: implement cli telemetry (#325) * feat: implement anonymous cli telemetry for run events * lock file * chore: bump deepagents to 1.11.0 to restore execute tool via routePrefixes fix (#357) * ci: fix repository posture checks * feat: Add trending repo badge to readme (#347) * chore(deps): bump the major group with 5 updates (#348) Bumps the major group with 5 updates: | Package | From | To | | --- | --- | --- | | [actions/checkout](https://github.com/actions/checkout) | `4` | `7` | | [pnpm/action-setup](https://github.com/pnpm/action-setup) | `4.3.0` | `6.0.9` | | [actions/setup-node](https://github.com/actions/setup-node) | `4` | `7` | | [github/codeql-action](https://github.com/github/codeql-action) | `3` | `4` | | [peter-evans/create-pull-request](https://github.com/peter-evans/create-pull-request) | `7.0.11` | `8.1.1` | Updates `actions/checkout` from 4 to 7 - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](actions/checkout@v4...v7) Updates `pnpm/action-setup` from 4.3.0 to 6.0.9 - [Release notes](https://github.com/pnpm/action-setup/releases) - [Commits](pnpm/action-setup@b906aff...0ebf471) Updates `actions/setup-node` from 4 to 7 - [Release notes](https://github.com/actions/setup-node/releases) - [Commits](actions/setup-node@v4...v7) Updates `github/codeql-action` from 3 to 4 - [Release notes](https://github.com/github/codeql-action/releases) - [Changelog](https://github.com/github/codeql-action/blob/main/CHANGELOG.md) - [Commits](github/codeql-action@v3...v4) Updates `peter-evans/create-pull-request` from 7.0.11 to 8.1.1 - [Release notes](https://github.com/peter-evans/create-pull-request/releases) - [Commits](peter-evans/create-pull-request@22a9089...5f6978f) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: pnpm/action-setup dependency-version: 6.0.9 dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: actions/setup-node dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: github/codeql-action dependency-version: '4' dependency-type: direct:production update-type: version-update:semver-major dependency-group: major - dependency-name: peter-evans/create-pull-request dependency-version: 8.1.1 dependency-type: direct:production update-type: version-update:semver-major dependency-group: major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Colin Francis <131073567+colifran@users.noreply.github.com> * feat: add gemini (ai studio) and gemini enterprise (vertex ai) providers with vertex model-family routing (#154) * feat: add Gemini providers and consolidate Vertex Claude into gemini-enterprise Add two Google model providers and route the Vertex AI (now Gemini Enterprise) Model Garden through a single provider: - `gemini` (AI Studio): Google's Gemini models via one API key (GEMINI_API_KEY). - `gemini-enterprise` (Vertex AI): keyless, ADC-authenticated provider that routes each model ID to the right Model Garden surface — native Gemini/Gemma over generateContent, Claude over the Anthropic Vertex SDK, and partner/open-weight models over the OpenAI-compatible MaaS endpoint (resolveVertexSurface / createGeminiEnterpriseModel). This supersedes the discrete `vertex` (Claude-only) provider added in #179: `gemini-enterprise` reaches Claude on Vertex through the same AnthropicVertex construction and ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN env-neutralization, plus publisher-path stripping, so no capability is lost. The rename tracks Google's Cloud Next 2026 rebrand of Vertex AI to the Gemini Enterprise Agent Platform while keeping the unchanged underlying API surface. Builds on #179's keyless-provider plumbing (projectEnvKey/locationEnvKey/ defaultLocation, getMissingProviderEnvKey, resolveProviderLocation, getProviderCredentialHint) and its gcp-project/gcp-location onboarding steps, which gemini-enterprise inherits — dropping #154's earlier approach of overloading apiKeyEnvKey to carry the non-secret GOOGLE_CLOUD_PROJECT. Gemini uses disableStreaming + outputVersion "v0" to preserve Gemini 3.x thought signatures across multi-turn tool calls (shared GEMINI_THOUGHT_SIGNATURE_OPTIONS across both ChatGoogle surfaces). The MaaS endpoint and the Gemini surface both special-case the `global` location: MaaS uses the unprefixed aiplatform.googleapis.com host, and the Gemini surface strips a publisher-pathed ID to the bare model the generateContent surface expects. resolveVertexSurface lists `codellama` explicitly so bare codellama-* IDs route to MaaS. Co-Authored-By: Claude <noreply@anthropic.com> * test: verify provider retry attempts reach the Gemini constructors createModel now spreads ...retryOptions into every Gemini surface (the AI Studio ChatGoogle and all three gemini-enterprise clients), which previously omitted it — so OPENWIKI_PROVIDER_RETRY_ATTEMPTS was silently ignored and a 429 failed fast instead of backing off (reported by @avidspartan1 on #154). maxRetries is not a readable property on the constructed LangChain models, so this asserts it via vi.mock capturing each constructor's options (including that maxRetries: 0 propagates and the AI Studio API key reaches ChatGoogle). Once maxRetries is wired through, @langchain/core's AsyncCaller already honors a server-provided Retry-After (up to 60s), so the existing backoff engages. Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> * bump deepagents to 1.11.0 to restore execute tool via routePrefixes fix --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: John Kennedy <65985482+jkennedyvz@users.noreply.github.com> Co-authored-by: Brace Sproul <braceasproul@gmail.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Brad Huffman <brad.huffman@wellsky.com> Co-authored-by: Claude <noreply@anthropic.com> * fix: Better relationship prompting (#359) * chore: Update openwiki to be OKF compliant (#355) * chore: Update openwiki to be OKF compliant * add index files * cr --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: Colin Francis <131073567+colifran@users.noreply.github.com> Co-authored-by: John Kennedy <65985482+jkennedyvz@users.noreply.github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Brad Huffman <brad.huffman@wellsky.com> Co-authored-by: Claude <noreply@anthropic.com>

SadriG91 force-pushed the feat/vertex-claude-provider branch 2 times, most recently from ad8ab66 to 6fd5afa Compare July 9, 2026 10:21

This was referenced Jul 9, 2026

feat: add Vertex AI provider for enterprise grade GCP integration #228

Closed

feat: add Vertex AI and Bedrock model providers #238

Closed

feat: offer provider-aware CI workflow scaffolding after --init #240

Open

Brad Huffman (ppsplus-bradh) mentioned this pull request Jul 12, 2026

feat: add gemini (ai studio) and gemini enterprise (vertex ai) providers with vertex model-family routing #154

Merged

SadriG91 force-pushed the feat/vertex-claude-provider branch from a06156b to 17d08ab Compare July 13, 2026 09:24

SadriG91 force-pushed the feat/vertex-claude-provider branch from 17d08ab to c44e272 Compare July 14, 2026 11:37

Colin Francis (colifran) changed the title ~~feat: add Google Vertex AI (Claude) provider~~ feat: add google vertex ai (claude) provider Jul 15, 2026

Colin Francis (colifran) reviewed Jul 15, 2026

View reviewed changes

SadriG91 added 3 commits July 15, 2026 13:25

fix: pass retryOptions to the vertex ChatAnthropic client

55d868f

SadriG91 force-pushed the feat/vertex-claude-provider branch from 4105b3e to 55d868f Compare July 15, 2026 11:26

Colin Francis (colifran) approved these changes Jul 15, 2026

View reviewed changes

Colin Francis (colifran) merged commit 15f1af5 into langchain-ai:main Jul 15, 2026
9 checks passed

This was referenced Jul 16, 2026

Scheduled OpenWiki updates run and fail on forks again after #179 #369

Closed

fix: keep scheduled OpenWiki updates opt-in on forks #370

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: add google vertex ai (claude) provider#179

feat: add google vertex ai (claude) provider#179
Colin Francis (colifran) merged 3 commits into
langchain-ai:mainfrom
SadriG91:feat/vertex-claude-provider

SadriG91 commented Jul 7, 2026

Uh oh!

SadriG91 commented Jul 7, 2026

Uh oh!

Brad Huffman (ppsplus-bradh) commented Jul 12, 2026

Uh oh!

SadriG91 commented Jul 13, 2026

Uh oh!

Colin Francis (colifran) commented Jul 15, 2026

Uh oh!

Colin Francis (colifran) Jul 15, 2026

Uh oh!

SadriG91 Jul 15, 2026 •

edited

Loading

Uh oh!

Colin Francis (colifran) left a comment

Uh oh!

SadriG91 commented Jul 15, 2026

Uh oh!

Colin Francis (colifran) left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

SadriG91 commented Jul 7, 2026

What

Why

How

How I tested it

Uh oh!

SadriG91 commented Jul 7, 2026

Uh oh!

Brad Huffman (ppsplus-bradh) commented Jul 12, 2026

Uh oh!

SadriG91 commented Jul 13, 2026

Uh oh!

Colin Francis (colifran) commented Jul 15, 2026

Uh oh!

Colin Francis (colifran) Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

SadriG91 Jul 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Colin Francis (colifran) left a comment

Choose a reason for hiding this comment

Uh oh!

SadriG91 commented Jul 15, 2026

Uh oh!

Colin Francis (colifran) left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

SadriG91 Jul 15, 2026 •

edited

Loading